More general patch for "mips offs > 16bit", (emit_insn/force_operand inconsistencies, expr.c 1.76).

[Jeffrey A Law]

  In message < Pine.BSF.4.02A.9903180234070.7478-100000@dair.pair.com >you write:
  > I did, and saw no error in it returning the (MEM (PLUS ...)).  An address
  > is not guaranteed to be an operand.  The construct "(plus reg cont_int)"
  > is normally a valid address, but never a valid operand to a move insn.
It is an error if the address is not valid.  As is the case if the integer
is large.


  > No (*sigh*), I believe you miss too see the difference between a valid
  > address and a valid operand here.
No.  I do see the difference.  A valid address should be a valid operand in
this case.

If it wasn't all kinds of things in the compiler will not work.

We may still want to put force_operand calls in later, but that would just
paper over the bug in the mips backend.

jeff